Running large DB's on FreeBSD

[Ronald Klop]

On Tue, 24 Oct 2006 02:00:22 +0200, Chuck Swiger <cswiger at mac.com> wrote:

> On Oct 23, 2006, at 4:44 PM, Mike Jakubik wrote:
>>> Moderately...it kinda depends on the budget available.  I regard  
>>> Solaris + Oracle as one of the most reliable combinations for moderate  
>>> to extreme load, for a system that might well be in operation for five  
>>> to ten years.  If I was going to do FreeBSD, I might look into  
>>> Postgres instead of MySQL; well, I might look into something else than  
>>> MySQL under many circumstances.  I've gotten some pretty good use out  
>>> of OpenBase, for another choice.
>>
>> I believe the front-end application is MySQL dependent, but what is so  
>> much better about PostgreSQL? I understand that it has some more  
>> advanced features, but if they are not used, then what is the  
>> advantage? (I really like the InnooDB storage in MySQL)
>
> I'm not sure whether avoiding deadlocks and using row-level locking by  
> default qualifies as "advanced features", but unless you use InnoDB with  
> MySQL, you don't get that from MySQL.  Postgres has been around for a  
> lot longer, and isn't as volatile as MySQL seems to be; also, it avoids  
> some of the needless timer overhead that MySQL seems to enjoy, and the  
> less-accurate-but-much-quicker gettimeofday() under Linux helps MySQL on  
> that platform versus FreeBSD.
>
>>> As for the disk configuration, using RAID-5 is one of the worst  
>>> possible choices for a database; using multiple RAID-1 mirrors or a  
>>> RAID-10 config would probably do a lot better in terms of performance  
>>> and reliability.
>>
>> Is RAID5 really that bad when a lot of fast disks are used and the  
>> controller has a decent cache with a BBWC? Thanks for the feedback guys.
>
> Yes, RAID-5 really can be that bad, unless your database is read-only or  
> read-mostly.  Lots of small writes will perform badly under RAID-5, even  
> with a battery-backed write-cache in write-back mode...

Example: writing 1 bit on 1 disk needs to read some info from all disks to  
recalculate the parity. So this doesn't scale very well.


-- 
  Ronald Klop
  Amsterdam, The Netherlands